The annual review as reconciliation.

The tell was in the calendar. Output climbed visibly in the weeks before review season and settled back afterwards, not out of cynicism, but because that is what an annual judgement rewards. If a whole year is going to be assessed in one conversation, being most visible immediately before that conversation is the rational strategy. I would have done the same.

Recency bias is a two-sided problem

The half that gets discussed is the smaller one. Yes, a manager remembers the last eight weeks most clearly. But the team, reasonably enough, optimises for those same eight weeks. Correcting only my own memory would have left the incentive completely intact; I’d have been a fairer judge of a game people were still playing.

So I moved the measurement to where the work actually happens, and left the review to do something else entirely.

What the calibration reads

The output record

I open the designer’s historical chart from the monthly velocity model: the volume and complexity of what they delivered, and how often they stepped up to absorb cover for a teammate. Twelve data points, all recorded at the time, none of them assembled in hindsight.

The behavioural record

I cross-reference that against their four-quarter balanced average from the peer framework. This is the part that removes my personal bias: it shows how reliably they held delivery timelines, precision and respectful communication with engineering, sales and marketing, as judged by those teams, not by me.

The conversation itself

With the numbers already settled and visible to both of us, the review session is free to be about something more useful than scoring. We talk about operational growth: did they expand their domain ownership? Did they turn a manual pipeline into an agent skill the rest of the team now uses? Did they protect the business from an escalation that never had to reach me?

What it changed

Because everything is recorded as it happens, the annual review stops being a verdict and becomes a reconciliation; we are reading a year of data we both already hold, not reconstructing it from memory. There is nothing to perform for in the fourth quarter, because the fourth quarter counts for exactly as much as the first.

That is what took the stress out of review season. It also removed the reason for the performance spike that started all of this: when effort is measured continuously, there is no window worth saving it for. The team knows how they are assessed in any given week rather than once a year, and that consistency is a large part of why nobody has left in two and a half years.

My job isn’t to make designers dependent on me

A manager can make a team look productive by answering every question, reviewing every decision and stepping in whenever something goes wrong. It works, briefly. It also does not survive contact with the manager’s calendar, and it produces designers who are very good at checking.

So I try to do the opposite. Give designers ownership. Give them room to make the decision. Take the organisational noise off their desk. Step in when they are genuinely stuck, blocked, or exposed to a problem they should not have to solve alone.

One of my product designers had come from a much more closely managed environment, where checking first was the safe move and had been the expected one. I deliberately handed him more ownership of his own decisions, and the room to build his own design judgement rather than learn to predict mine. That did not mean disappearing. I stayed close enough to unblock him, to challenge a call when it needed challenging, and to keep the work protected when organisational politics started leaning on it.

What came back was not just a designer with more freedom. It was better work, and someone operating a level above where he started without a manager standing next to him. That is the direction I want every one of these conversations to go: the designer needs me less for decisions, and more for the problems only a manager can solve.

Five systems, one habit each

None of these started as a plan for a system. Each was built the week a specific habit stopped working, which is why they interlock: every later one assumes the earlier one exists. The repository holds the record. The ownership matrix decides where work lands. Intake decides how it arrives. Velocity and peer review record what happened. And this one simply reads it all back.